ORYNX

Cline, Roo Code and Kilo Code

Configure the VS Code coding agents to use ORYNX.

Cline and Roo Code share the same provider settings. Kilo Code adds ORYNX as a custom provider.

Roo Code was discontinued in May 2026. If you still run its last release, follow the Cline steps.

Cline and Roo Code

OpenAI Compatible provider

This works for every model except Gemini, which ORYNX serves only in the Gemini format.

  1. Open the extension settings and set API Provider to OpenAI Compatible.
  2. Base URL: https://api.orynx.dev/v1
  3. API Key: your sk-orynx-... key
  4. Model: pick one from the list loaded from ORYNX, or type an id such as kimi-k3, glm-5.3 or claude-sonnet-5.
  5. Set the context window, max output tokens and image support to the values on the model's page under Models. The extensions cannot read them from ORYNX.

Anthropic provider for Claude

Use the Anthropic provider for Claude's prompt caching and extended thinking:

  1. Set API Provider to Anthropic and paste your key into the API key field.
  2. Turn on Use custom base URL and enter https://api.orynx.dev (no /v1).
  3. Pick a Claude model.

Roo Code's Anthropic provider only lists the Claude models it shipped with, up to Sonnet 4.6 and Opus 4.6. For newer Claude models in Roo Code, use the OpenAI Compatible provider.

Kilo Code

  1. Open Settings (gear icon), go to the Providers tab and click Custom provider at the bottom.
  2. Fill in the dialog:
    • Provider ID: orynx
    • Display name: ORYNX
    • Provider API: OpenAI Compatible
    • Base URL: https://api.orynx.dev/v1
    • API key: your sk-orynx-... key
    • Models: pick from the list Kilo loads from ORYNX, or add ids by hand
  3. Click Submit.

For Claude in the Anthropic format, add a second custom provider (for example orynx-claude) with Provider API set to Anthropic Messages and the same base URL, https://api.orynx.dev/v1. Kilo's Anthropic client adds only /messages, so this base URL needs /v1.

The dialog does not set token limits. When Kilo cannot find a model id in its built-in catalog, the limits fall back to 0 and automatic context compaction is off. Add a limit to each model of the provider in ~/.config/kilo/kilo.jsonc, using the numbers on the model's page:

~/.config/kilo/kilo.jsonc
{
  "provider": {
    "orynx": {
      "models": {
        "kimi-k3": {
          "name": "Kimi K3",
          "limit": { "context": 1048576, "output": 131072 }
        }
      }
    }
  }
}

On this page